Search

Summary: Background: Deep learning offers considerable promise for medical diagnostics. We aimed to evaluate the diagnostic accuracy of deep learning algorithms versus health-care professionals in classifying diseases using medical imaging. Methods: In this systematic review and meta-analysis, we searched Ovid-MEDLINE, Embase, Science Citation Index, and Conference Proceedings Citation Index for studies published from Jan 1, 2012, to June 6, 2019. Studies comparing the diagnostic performance of deep learning models and health-care professionals based on medical imaging, for any disease, were included. We excluded studies that used medical waveform data graphics material or investigated the accuracy of image segmentation rather than disease classification. We extracted binary diagnostic accuracy data and constructed contingency tables to derive the outcomes of interest: sensitivity and specificity. Studies undertaking an out-of-sample external validation were included in a meta-analysis, using a unified hierarchical model. This study is registered with PROSPERO, CRD42018091176. Findings: Our search identified 31 587 studies, of which 82 (describing 147 patient cohorts) were included. 69 studies provided enough data to construct contingency tables, enabling calculation of test accuracy, with sensitivity ranging from 9·7% to 100·0% (mean 79·1%, SD 0·2) and specificity ranging from 38·9% to 100·0% (mean 88·3%, SD 0·1). An out-of-sample external validation was done in 25 studies, of which 14 made the comparison between deep learning models and health-care professionals in the same sample. Comparison of the performance between health-care professionals in these 14 studies, when restricting the analysis to the contingency table for each study reporting the highest accuracy, found a pooled sensitivity of 87·0% (95% CI 83·0–90·2) for deep learning models and 86·4% (79·9–91·0) for health-care professionals, and a pooled specificity of 92·5% (95% CI 85·1–96·4) for deep learning models and 90·5% (80·6–95·7) for health-care professionals. Interpretation: Our review found the diagnostic performance of deep learning models to be equivalent to that of health-care professionals. However, a major finding of the review is that few studies presented externally validated results or compared the performance of deep learning models and health-care professionals using the same sample. Additionally, poor reporting is prevalent in deep learning studies, which limits reliable interpretation of the reported diagnostic accuracy. New reporting standards that address specific challenges of deep learning could improve future studies, enabling greater confidence in the results of future evaluations of this promising technology. Funding: None.

Summary: Background: Deep learning has the potential to transform health care; however, substantial expertise is required to train such models. We sought to evaluate the utility of automated deep learning software to develop medical image diagnostic classifiers by health-care professionals with no coding—and no deep learning—expertise. Methods: We used five publicly available open-source datasets: retinal fundus images (MESSIDOR); optical coherence tomography (OCT) images (Guangzhou Medical University and Shiley Eye Institute, version 3); images of skin lesions (Human Against Machine [HAM] 10000), and both paediatric and adult chest x-ray (CXR) images (Guangzhou Medical University and Shiley Eye Institute, version 3 and the National Institute of Health [NIH] dataset, respectively) to separately feed into a neural architecture search framework, hosted through Google Cloud AutoML, that automatically developed a deep learning architecture to classify common diseases. Sensitivity (recall), specificity, and positive predictive value (precision) were used to evaluate the diagnostic properties of the models. The discriminative performance was assessed using the area under the precision recall curve (AUPRC). In the case of the deep learning model developed on a subset of the HAM10000 dataset, we did external validation using the Edinburgh Dermofit Library dataset. Findings: Diagnostic properties and discriminative performance from internal validations were high in the binary classification tasks (sensitivity 73·3–97·0%; specificity 67–100%; AUPRC 0·87–1·00). In the multiple classification tasks, the diagnostic properties ranged from 38% to 100% for sensitivity and from 67% to 100% for specificity. The discriminative performance in terms of AUPRC ranged from 0·57 to 1·00 in the five automated deep learning models. In an external validation using the Edinburgh Dermofit Library dataset, the automated deep learning model showed an AUPRC of 0·47, with a sensitivity of 49% and a positive predictive value of 52%. Interpretation: All models, except the automated deep learning model trained on the multilabel classification task of the NIH CXR14 dataset, showed comparable discriminative performance and diagnostic properties to state-of-the-art performing deep learning algorithms. The performance in the external validation study was low. The quality of the open-access datasets (including insufficient information about patient flow and demographics) and the absence of measurement for precision, such as confidence intervals, constituted the major limitations of this study. The availability of automated deep learning platforms provide an opportunity for the medical community to enhance their understanding in model development and evaluation. Although the derivation of classification models without requiring a deep understanding of the mathematical, statistical, and programming principles is attractive, comparable performance to expertly designed models is limited to more elementary classification tasks. Furthermore, care should be placed in adhering to ethical principles when using these automated models to avoid discrimination and causing harm. Future studies should compare several application programming interfaces on thoroughly curated datasets. Funding: National Institute for Health Research and Moorfields Eye Charity.

Purpose: Application of artificial intelligence (AI) to macular OCT scans to segment and quantify volumetric change in anatomical and pathological features during intravitreal treatment for neovascular age-related macular degeneration (AMD). Design: Retrospective analysis of OCT images from the Moorfields Eye Hospital AMD Database. Participants: A total of 2115 eyes from 1801 patients starting anti-VEGF treatment between June 1, 2012, and June 30, 2017. Methods: The Moorfields Eye Hospital neovascular AMD database was queried for first and second eyes receiving anti-VEGF treatment and had an OCT scan at baseline and 12 months. Follow-up scans were input into the AI system and volumes of OCT variables were studied at different time points and compared with baseline volume groups. Cross-sectional comparisons between time points were conducted using Mann–Whitney U test. Main Outcome Measures: Volume outputs of the following variables were studied: intraretinal fluid, subretinal fluid, pigment epithelial detachment (PED), subretinal hyperreflective material (SHRM), hyperreflective foci, neurosensory retina, and retinal pigment epithelium. Results: Mean volumes of analyzed features decreased significantly from baseline to both 4 and 12 months, in both first-treated and second-treated eyes. Pathological features that reflect exudation, including pure fluid components (intraretinal fluid and subretinal fluid) and those with fluid and fibrovascular tissue (PED and SHRM), displayed similar responses to treatment over 12 months. Mean PED and SHRM volumes showed less pronounced but also substantial decreases over the first 2 months, reaching a plateau postloading phase, and minimal change to 12 months. Both neurosensory retina and retinal pigment epithelium volumes showed gradual reductions over time, and were not as substantial as exudative features. Conclusions: We report the results of a quantitative analysis of change in retinal segmented features over time, enabled by an AI segmentation system. Cross-sectional analysis at multiple time points demonstrated significant associations between baseline OCT-derived segmented features and the volume of biomarkers at follow-up. Demonstrating how certain OCT biomarkers progress with treatment and the impact of pretreatment retinal morphology on different structural volumes may provide novel insights into disease mechanisms and aid the personalization of care. Data will be made public for future studies. Financial Disclosure(s): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Purpose: Rare disease diagnosis is challenging in medical image-based artificial intelligence due to a natural class imbalance in datasets, leading to biased prediction models. Inherited retinal diseases (IRDs) are a research domain that particularly faces this issue. This study investigates the applicability of synthetic data in improving artificial intelligence-enabled diagnosis of IRDs using generative adversarial networks (GANs). Design: Diagnostic study of gene-labeled fundus autofluorescence (FAF) IRD images using deep learning. Participants: Moorfields Eye Hospital (MEH) dataset of 15 692 FAF images obtained from 1800 patients with confirmed genetic diagnosis of 1 of 36 IRD genes. Methods: A StyleGAN2 model is trained on the IRD dataset to generate 512 × 512 resolution images. Convolutional neural networks are trained for classification using different synthetically augmented datasets, including real IRD images plus 1800 and 3600 synthetic images, and a fully rebalanced dataset. We also perform an experiment with only synthetic data. All models are compared against a baseline convolutional neural network trained only on real data. Main Outcome Measures: We evaluated synthetic data quality using a Visual Turing Test conducted with 4 ophthalmologists from MEH. Synthetic and real images were compared using feature space visualization, similarity analysis to detect memorized images, and Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) score for no-reference-based quality evaluation. Convolutional neural network diagnostic performance was determined on a held-out test set using the area under the receiver operating characteristic curve (AUROC) and Cohen’s Kappa (κ). Results: An average true recognition rate of 63% and fake recognition rate of 47% was obtained from the Visual Turing Test. Thus, a considerable proportion of the synthetic images were classified as real by clinical experts. Similarity analysis showed that the synthetic images were not copies of the real images, indicating that copied real images, meaning the GAN was able to generalize. However, BRISQUE score analysis indicated that synthetic images were of significantly lower quality overall than real images (P < 0.05). Comparing the rebalanced model (RB) with the baseline (R), no significant change in the average AUROC and κ was found (R-AUROC = 0.86[0.85-88], RB-AUROC = 0.88[0.86-0.89], R-κ = 0.51[0.49-0.53], and RB-κ = 0.52[0.50-0.54]). The synthetic data trained model (S) achieved similar performance as the baseline (S-AUROC = 0.86[0.85-87], S-κ = 0.48[0.46-0.50]). Conclusions: Synthetic generation of realistic IRD FAF images is feasible. Synthetic data augmentation does not deliver improvements in classification performance. However, synthetic data alone deliver a similar performance as real data, and hence may be useful as a proxy to real data.Financial Disclosure(s): Proprietary or commercial disclosure may be found after the references.